Re: [PATCH v9 00/14] Add support for remote unwind

[Hekuang]

hi

å 2016/6/8 3:44, Arnaldo Carvalho de Melo åé:
> Em Fri, Jun 03, 2016 at 09:06:29AM +0200, Jiri Olsa escreveu:
> > On Fri, Jun 03, 2016 at 03:33:09AM +0000, He Kuang wrote:
> >
> > SNIP
> >
> > > For using remote libunwind libraries, reference this:
> > >    http://thread.gmane.org/gmane.linux.kernel/2224430
> > >
> > > and now we can use LIBUNWIND_DIR to specific custom dirctories
> > > containing libunwind libs.
> > >
> > > Acked-by: Jiri Olsa <jolsa@xxxxxxxxxx> for most patches except:
> > >
> > > v9:
> > >   - Change function unwind__register_ops() to static.
> > >   - Move up unwind__prepare_access() in thread__insert_map() and save
> > >     map_groups__remove() call.
> > >   - Enclose multiple line if/else into braces.
> > >   - Fix miss modified function declaration for unwind__prepare_access()
> > >     in patch 10.
> > for patchset:
> >
> > Acked-by: Jiri Olsa <jolsa@xxxxxxxxxx>
> Ok, I'm applying it, after fixing 'perf test unwind', 'perf top --call-graph dwarf'
> and 'perf trace --call-graph dwarf', but I have one question, is the
> scenario where we collect on a x86_64 machine and want to do analysis on
> a ARM64 or x86-32 machine supported? This should be the odd case now,
Yes, it's supported.

But I never tested this before, so I just compiled libunwind for
aarch64, and tested unwinding i686 perf.data on aarch64. Then I
found another issue I've considered but missed at some version of
this patch series.

In util/unwind-libunwind-local.c, PERF_REG_SP/IP is used, but
those macros are assigned to the host platform, we should
redefine them in the wrapper file, for example in
"util/libunwind/x86_32.c".

After fixing this problem, i686 perf.data can be parsed on
aarch64 machine.  Since you've already applied the v9 patches,
should I send patches based on the lastest tree as bug fixes or
just update v9 patches?

Thank you.

Here is the modified part:

diff --git a/tools/perf/util/libunwind/arm64.c 
b/tools/perf/util/libunwind/arm64.c
index 4fb5395..8a5c2fc 100644
--- a/tools/perf/util/libunwind/arm64.c
+++ b/tools/perf/util/libunwind/arm64.c
@@ -29,6 +29,11 @@
 #ifdef NO_LIBUNWIND_DEBUG_FRAME_AARCH64
 #define NO_LIBUNWIND_DEBUG_FRAME
 #endif
+
+#undef PERF_REG_IP
+#undef PERF_REG_SP
+#define PERF_REG_IP PERF_REG_ARM64_PC
+#define PERF_REG_SP PERF_REG_ARM64_SP
 #include "util/unwind-libunwind-local.c"

 struct unwind_libunwind_ops *
diff --git a/tools/perf/util/libunwind/x86_32.c 
b/tools/perf/util/libunwind/x86_32.c
index d98c17e..de21a39 100644
--- a/tools/perf/util/libunwind/x86_32.c
+++ b/tools/perf/util/libunwind/x86_32.c
@@ -31,6 +31,11 @@
 #ifndef NO_LIBUNWIND_DEBUG_FRAME
 #define NO_LIBUNWIND_DEBUG_FRAME
 #endif
+
+#undef PERF_REG_IP
+#undef PERF_REG_SP
+#define PERF_REG_IP PERF_REG_X86_IP
+#define PERF_REG_SP PERF_REG_X86_SP
 #include "util/unwind-libunwind-local.c"

 struct unwind_libunwind_ops *


> but from a quick look I couldn't see this as being supported, is that
> true or I was just lazy not to have tried this?
>
> - Arnaldo